Stress Testing-Principles Practice, MYRVIN H. ELLESTAD, fifth edition

278 STRESS TESTING: PRINCIPLES AND PRACTICE uate the probability of disease.3 Both pretest and post-test probability will be calculated after a stress test or a series of noninvasive tests including cardiokymography, coronary calcification by fluoroscopy, and thallium scintigraphy. Hossack and colleagues5 believe one can do as well with the conven- tional risk factors combined with exercise risk factors obtained during tread- mill stress testing. The Framingham risk factor tables are well known and have been used by the Seattle Group in their Heart Watch study for a num- ber of years. The use of symptoms for disease prevalence presumes that the patient will tell you about pain. Many subjects withhold or modify informa- tion for various reasons; commercial airline pilots are a typical example. EFFECT OF PREVALENCE ON EXERCISE-INDUCED ST DEPRESSION To proceed with the application of Bayes’ theorem, the probability of signifi- cant CAD is presented from Diamond’s calculations based on the pretest prob- ability. Figure 14–5 illustrates the degree of diagnostic uncertainty according to the magnitude of ST-segment depression and the pretest probability. The data suggest that even if the pretest probability was only 20%, exercise-induced ST depression of greater than 2.5 mm gives a probability of disease in the range of 90%, whereas slightly less than 2 mm of ST depres- sion results in a probability of about 50%. This information can be used after the exercise test and in other tests such as a nuclear scintigram. For example, if the post-test probability is 70% after a stress test and the information con- tent of a nuclear test is 25%, an abnormal nuclear scintigram will give a prob- ability of 95%, however, a normal nuclear test following the abnormal stress test will reduce the probability to 55%. FIGURE 14–5. Family of ST-segment depression curves and the likelihood of CAD. (From Epstein,6 with permission.)

PREDICTIVE IMPLICATIONS 279 CRITIQUE OF BAYES’ THEOREM The numbers presented here provide a highly simplified approach to a com- plex problem. Although the concept is valid, when we apply it to our patients we must remember that important elements in the calculations are not actu- ally known with certainty. The sensitivity and specificity of stress testing in the day-to-day man- agement of particular patients are uncertain. Most of the data available to us come from cardiac centers where the referral pattern may influence the prevalence of disease in the study.7 Sensitivity and specificity from such cen- ters are known to vary. How then do we relate these data to our individual patients? In clinical practice, most of us instinctively use a number of vari- ables to determine the presence of disease. In some laboratories, a computer- generated probability based on a multivariate analysis of a number of vari- ables is used. However, we have no information that indicates how well this approach would work in an outpatient-oriented clinic, a private practice of cardiology, or an internal medical office. POPULATION GROUPS In a cardiology practice in an area where prevalence of CAD is very high (es- pecially if the physician is a recognized expert in this field), there would be a larger percentage of patients with CAD than in the office of a general in- ternist, in an industrial clinic, or in a military installation. An example of this concept is illustrated in Table 14–5, in which two groups of subjects were studied in relationship to the prevalence of exercise-induced ST depression (positive stress tests). Those referred for evaluation in the Memorial Hospital Cardiology Lab- oratory, as expected, resulted in two to four times more positive tests than those studied in an asymptomatic group solicited by the Long Beach Heart Association.8 Actually, most physicians do stress tests on subjects who, by their age and sex, are in a population with a higher prevalence of CAD. Table 14–5. Percent of Positive Stress Tests According to Age and Sex in Two Studies Age MHLB* Female LBHA† Female MHLB Male LBHA Male 21–30 0 0 2.5 1.2 31–40 10.1 2.1 11.7 4.3 41–50 19.9 2.0 29.5 11.4 51–60 29.1 7.4 48.0 26.9 Over 60 43.3 12.1 58.2 29.3 Mean 23.3 4.6 34.3 13.5 *Memorial Hospital Cardiology Laboratory. †Long Beach Heart Association. From Ellestad, Allen, and Stuart, with permission.46

280 STRESS TESTING: PRINCIPLES AND PRACTICE Using Table 14–5, it would be fair to estimate a disease prevalence of approximately 19% (26.9 + 11.4 = 19) Ϭ 2 in men between ages 40 and 60. In this population, there would be a significant number of false-positives and false-negatives. If we were to agree that about 20% were false-posi- tives, we might calculate the predictive value if we analyzed 500 men as follows: No. with abnormal test No. with normal test 100 diseased 70 (TP) 30 FN (Sensitivity = 70%) 400 nondiseased 80 (FP) 320 TN (Specificity = 80%) 350 Total 150 Predicted value of abnormal test = ᎏTPT+PFP = ᎏ17500 = 46% where TP = true-positive; TN = true-negative; FP = false-positive; and FN ϭ false-negative. Thus, almost 50% of this group of men would be correctly identified by ST segments and age alone. It would be prudent to evaluate the abnormal responders by considering risk factors and other exercise variables (see Chapter 19). LIMITED CHALLENGE When evaluating patients to determine the accuracy of the test some inves- tigators have compared patients with a high probability of disease, i.e. pa- tients with chest pain, with a group who obviously have a low probability, such as normal volunteers. This has been termed “limited challenge” because it exaggerates the tendency for normal patients to have a negative test and thus increase the specificity. The best way to know the usefulness of the test is to compare only patients who are suspected of ischemia, usually because they present with chest pain. The final caveat has to do with the difficulties with angiographic es- timates of coronary narrowing and their effect on coronary flow during metabolically induced hyperemia.9 When we consider that all our data, using angiographically estimated percentage of narrowing, are subject to considerable question (see Chapter 4), it may take a few years before the reliability of any test in estimating coronary ischemia can be based on more certain criteria. For a more detailed critique of this subject, I sug- gest that the serious reader review Feinstein’s10 essay, ”The Haze of Bayes, the Aerial Palaces of Decision Analysis, and the Computerized Ouija Board.”